One-Class Support Vector Machine for WiFi-based Device-free Indoor Presence Detection

The utilization of existing radio signals such as 802.11 (WiFi) for device-free detection of human presence and movement indoors has garnered significant interest among researchers in academia and industry. Improving the efficiency of buildings, particularly in terms of heating and energy costs, relies on accurately detecting room occupancy. Our approach uses channel state information (CSI) obtained from commodity 802.11ac hardware as input to machine learning based on One Class Support Vector Machine (OC-SVM). Unlike other methods that necessitate extensive learning in environments with and without human presence, our approach treats human presence as a novelty. This simplifies the training process, as we only need to learn from environments without human presence, specifically empty rooms. Furthermore, since we focus solely on analyzing the magnitude information of the CSI data, there is no requirement for intricate sanitization of the phase information. Experimental results using standard WiFi hardware demonstrate exceptional performance, with accuracy, sensitivity, and specificity exceeding 97% in most cases. Furthermore, our proposed approach is practical, as it incurs minimal overhead in terms of radio resource usage. Simply capturing CSI data with a sampling rate of 5 Hz on only a few OFDM subcarriers from a 5MHz channel is sufficient.